OWLDAvis

[ajdapretnar]

Issue

Widget for LDA topic model visualization, inspired by pyLDAvis: https://github.com/bmabey/pyLDAvis

Description of changes

Add Relevant Terms widget for interactive exploration of relevant terms per topic.

Includes

• Code changes
• Tests
• Documentation

[ajdapretnar]

TODO:

• test the widget (GUI)
• test the method (should yield same results as pyLDAvis)
• implement bar charts instead of Distributions column (remove weights, too)
• documentation
• tests
• visual settings

[ajdapretnar]

Here's how the widget would look like. The distribution bars would be red (current gray) and blue (current white).

[ajdapretnar]

To discuss:

• display only words, not POS tags (actually, there's no way around it as the model should distinguish between work_NOUN and work_VERB)
• check weights, because they seem wrong (they are also correct since they push up the topic_prob:marginal_prob for more frequent topics - could be an optional parameter)

[codecov-commenter]

Codecov Report

Merging #731 (2eefd5c) into master (ee5709e) will increase coverage by 0.37%.
The diff coverage is 89.20%.

❗ Current head 2eefd5c differs from pull request most recent head fea8f68. Consider uploading reports for the commit fea8f68 to get more accurate results

@@            Coverage Diff             @@
##           master     #731      +/-   ##
==========================================
+ Coverage   74.24%   74.62%   +0.37%     
==========================================
  Files          72       74       +2     
  Lines        9591     9872     +281     
  Branches     1307     1331      +24     
==========================================
+ Hits         7121     7367     +246     
- Misses       2222     2250      +28     
- Partials      248      255       +7